[ZODB-Dev] ReadConflictError while re-indexing big fat index

[sean.upton at uniontrib.com]

Tim Peters wrote:
>     you read object 1
>     another transaction commits a change to object 2
>     you get an invalidation for object 2 as a result
>     you try to read object 2
> 
> Now you get a read conflict:  ZODB is protecting you from 
> seeing object 2 in a state that may be inconsistent with the 
> state you already have for object 1.  If you had read object 
> 2 before it was invalidated, no problem, then the states you 
> see for objects 1 and 2 are consistent.

So is there any way to re-read the object based on it's new state after you
catch the RCE and want to retry (outside the publisher machinery)?  I'm
thinking something less complicated than "re-loading" an entire index.
Let's say, just a simple content object gets invalidated.  What's the "right
way" to reload in a scenario like this:

app = Zope.app()
retries = 0
while 1:
    try:
        foo = app['foo'].__of__(app)
        #another thread/client changes app.foo here
        print foo.bar #raises RCE
        break
    except ReadConflictError:
        retries += 1
        #do something here to make the air safe to breath again?
        if retries > 4: break

Is there any way to get the new, consistent state for app.foo in the except
block above, so that I'm not retrying in vain?  Will get_connection().sync()
do the trick or is there something more?

This stuff is painful - I think I will be jumping on the 2.8/MVCC testing
bandwagon soon with some test instances of our production apps.

Thanks,
Sean

+----------------------------------------------------------+
 Sean Upton                              SignOnSanDiego.com 
 Site Technology Supervisor     The San Diego Union-Tribune 
 619.718.5241                 sean.upton at signonsandiego.com 
 2375 Northside Dr #300                 San Diego, CA 92108

  Plone Powered!  plone.org  ++  python.org  ++  zope.org   
+----------------------------------------------------------+